Google Search Console: The New Era of AI Visibility in France

The Search Console, Google’s Only Official Tool for AI

Since June 3, 2026, Google has introduced a specific report to evaluate the performance of websites in the context of generative AI, accessible via the Search Console. This initiative was announced by Hillel Maoz, head of engineering for the search ecosystem, and Moshe Samet, senior product manager of the Search Console. For the first time, this report provides a distinct view of site visibility within generative search features. In France, this report has been available since early July, marking a significant advancement for French webmasters.

What the Report Displays

The Search Console report offers a separate view for searches, which includes AI Overviews and AI Mode, as well as another for Discover. It focuses on a primary metric: impressions. This metric is broken down into several dimensions, including:

• The number of impressions, meaning how many times a site's URLs appear in an AI feature.
• The pages involved, grouped by canonical URL, allowing visibility into which pages are most prominent.
• The countries of origin for searches, providing a geographical perspective on visibility.
• The devices used, but only for the search view, which helps understand on which type of device the site is most accessed.
• Dates, with hourly granularity, allowing for detailed temporal analysis.

What the Report Does Not Display

Despite these features, the report has notable limitations. It does not provide the number of clicks, click-through rates, average position, or a breakdown by query. In other words, it measures a site's exposure—i.e., being mentioned—without indicating how many visits this exposure actually generates. A high volume of impressions could coincide with a decrease in traffic, as users often get their answers directly without clicking. However, Google has hinted that additional metrics may be added later.

Third-Party Tools and Manual Testing to Complete the Picture

The Search Console, while useful, only covers Google's surfaces and is limited to impressions. To obtain a more comprehensive overview, there are two supplements: specialized platforms and manual testing.

Tracking Platforms

There are primarily two types of tracking platforms. On one hand, SEO suites that have integrated an AI visibility module, such as Semrush with its AI Toolkit, Ahrefs with its Brand Radar, or SE Ranking. Other players like HubSpot, Semactic, Brand Score, GetMint, Meteoria, or Hikoo also offer tools that cover up to a dozen engines, including AI Mode and AI Overviews.

These tools do not all measure the same aspects. Some focus on mentions, when the brand is named, while others focus on citations, when a link points to the site. Others still measure share of voice compared to competitors or the sentiment of the mention. It is therefore crucial to choose based on the intended goal rather than the number of engines covered.

Manual Testing, Free and Immediate

The simplest method, although not always the most reliable, is to enter strategic queries into Google and observe if the site is mentioned in the AI Overview or AI Mode. This method is free and can be easily repeated, but it should be automated to avoid consuming too much time. However, the responses from the models can vary, meaning that an isolated test is not a reliable measure. A generated response may include your site or brand, while a similar query a few minutes later might not.

Conditions for Reliable Manual Testing

To ensure the reliability of manual tests, it is advisable to:

• Log out of your Google account or use a private browsing window to avoid personalization biases.
• Neutralize location when possible to obtain more neutral results.
• Vary the phrasing of the same query to test different ways the site might be mentioned.
• Repeat each test multiple times to check the consistency of results.
• Renew measurements weekly rather than monthly to track frequent variations.

Best Practices and Common Mistakes

Measuring AI visibility involves several pitfalls that can lead to erroneous conclusions. Here are some common mistakes to avoid:

• Confusing audience analysis with AI Overviews: a click from an AI Overview arrives with a google.com referrer and blends into organic traffic. Google Analytics 4 (GA4) does not isolate it. Its "AI Assistant" channel, rolled out in May 2026 and expanded in June, measures traffic from chatbots like ChatGPT, Gemini, DeepSeek, Copilot, and Grok, but not Perplexity, and does not account for traffic from AI Overviews.

• Taking AI traffic from GA4 as a total: a share between 35% and 70% of sessions referred by AIs arrives without a referrer and falls into direct traffic. Therefore, any figure obtained remains a floor.

• Testing from your own account: a test launched from a logged-in account and a usual location returns a biased result. Under neutral conditions, non-personalized profile, normalized location, and repeated passages, a site deemed absent often reappears in several responses.

• Measuring too infrequently: the stability of cited sources varies greatly from one surface to another. According to SISTRIX, out of 82,619 prompts tracked over seventeen weeks, AI Overviews remain stable for just over half of queries, while AI Mode renews 56% of its sources each week and ChatGPT up to 74%. On these shifting surfaces, monthly tracking captures only noise.

• Reading a score as a fixed position: model responses are not deterministic. A visibility score should be interpreted as a trend, not as a stable ranking.

• Relying solely on tracking tools: most query the models via their API, not through the engine actually used by the public, which introduces a gap with the real experience.

An important nuance completes this picture. Being cited says nothing about how the brand is presented. A study by Demand-Genius, conducted on ChatGPT in a B2B context, observes that with constant visibility, a model can steer its recommendation towards a competitor depending on the angle of the question, a finding that holds for conversational surfaces like AI Mode. Therefore, a score that progresses may mask an unfavorable recommendation.

The right approach is to cross-reference sources rather than rely on a single one. The Search Console report for exposure in Google’s AI, audience analysis for chatbot traffic, a manual test under neutral conditions, and, if the budget allows, a tracking platform. The current period, marked by a still partial rollout, is the right time to establish a benchmark before these responses become widespread.

An Incomplete Measurement at the Time of the French Rollout

The rollout of AI Overviews and AI Mode is taking place gradually in France, initially on a subset of queries and users. The Search Console report will thus begin to provide data specific to the French market, which until now has been mixed with other languages. Meanwhile, Google has equipped the Search Console with a button allowing a site to opt out of appearing in AI features. Sites that make this choice will not appear in these new features, which may have implications for their overall visibility.